iscc-bio - ISCC Processing for Bioimage Data

ISCC Processing for Multi-Dimensional Bioimage Data

Generate ISO 24138:2024 International Standard Content Codes (ISCC) for bioimage data across multiple formats using deterministic IMAGEWALK plane traversal.

Project Status

Version 0.2.0

[!WARNING] This package is a proof of concept, and breaking changes may be released at any time.

Overview

iscc-bio bridges bioimage formats with ISCC-CODE processing by implementing the IMAGEWALK specification - a deterministic algorithm for traversing and canonicalizing pixel data from multi-dimensional bioimaging data. This produces consistent, reproducible content identifiers regardless of source format or storage platform.

Documentation: https://bio.iscc.codes

Key Features

• Format-Agnostic Hashing: Generate reproducible ISCCs at the level of pixel data across OME-TIFF, OME-Zarr, OMERO, CZI, ND2, LIF, and other formats
• IMAGEWALK Implementation: Deterministic Z→C→T plane traversal with canonical byte representation
• Multi-Source Support: Process local files (via BioIO), OME-Zarr archives, and OMERO remote servers
• Memory Efficient: Lazy loading with Dask for processing large multi-dimensional images
• Multi-Scene Processing: Handle complex multi-scene/multi-series bioimage files
• Command-Line Tools: CLI commands for code generation

Installation

Basic Installation

# Using uv (recommended)
uv tool install iscc-bio

# Using pip
pip install iscc-bio

Installation with Format Support

# Install with all bioimage reader plugins
uv tool install "iscc-bio[readers]"

# Install with specific format support
uv tool install "iscc-bio[czi,nd2,lif]"

# Install everything (all readers)
uv tool install "iscc-bio[all]"

OMERO Support

OMERO requires platform-specific prebuilt zeroc-ice wheels not available on PyPI. Install separately:

pip install -r requirements-omero.txt

Available Optional Dependencies

• readers / all: All BioIO reader plugins (BioFormats, CZI, OME-TIFF, OME-Zarr, ND2, LIF, etc.)
• bioformats: BioFormats reader for broad format support
• czi, nd2, lif, ome-tiff, ome-zarr, dv, imageio, tifffile, tiff-glob: Individual format readers

Quick Start

CLI Commands

Generate Biocode (ISCC-SUM)

# Generate biocode (ISCC-SUM) for bioimage scenes
iscc-bio biocode myimage.ome.tiff

# Works with multiple sources:
iscc-bio biocode local/file.czi           # Local bioimage file
iscc-bio biocode data.zarr                # OME-Zarr/NGFF
iscc-bio biocode --host omero.server.com --iid 123  # OMERO server

# With per-plane simprints for similarity search:
iscc-bio biocode myimage.czi --simprints

Generate Imagecode (Experimental)

# Generate comprehensive fingerprint with ISCC-SUM + ISCC-IMAGE + ISCC-MIXED
iscc-bio imagecode myimage.czi

# Output includes:
# - ISCC-SUM hash over normalized pixel content
# - Representative view extraction (~5 views per scene)
# - ISCC-IMAGE codes for each view
# - ISCC-MIXED global descriptor

Extract Representative Views

# Extract intelligent 2D views for perceptual hashing
iscc-bio views myimage.nd2 --output-dir ./views/

# Extraction strategies:
# - Maximum intensity projections (MIP)
# - Best focus planes
# - Representative sampling
# - Multi-channel composites

IMAGEWALK Specification

IMAGEWALK is a deterministic algorithm for traversing multi-dimensional bioimage data to produce format-agnostic, reproducible hash digests. The canonical specification is maintained as ISCC Enhancement Proposal IEP-0018.

Core Principles

1. Z→C→T Traversal Order: Planes are processed in deterministic order:

   • Outermost loop: Z dimension (depth/focal plane)
   • Middle loop: C dimension (channel)
   • Innermost loop: T dimension (time)

2. Canonical Byte Representation: Each 2D plane is:

   • Flattened in row-major order (Y then X)
   • Encoded as big-endian bytes
   • Fed to a hash processor

3. Multi-Scene Independence: Each scene/series is processed separately, producing one hash per scene

Example Traversal

For an image with Z=2, C=3, T=2 dimensions (12 total planes):

Plane 1:  z=0, c=0, t=0    Plane 7:  z=1, c=0, t=0
Plane 2:  z=0, c=0, t=1    Plane 8:  z=1, c=0, t=1
Plane 3:  z=0, c=1, t=0    Plane 9:  z=1, c=1, t=0
Plane 4:  z=0, c=1, t=1    Plane 10: z=1, c=1, t=1
Plane 5:  z=0, c=2, t=0    Plane 11: z=1, c=2, t=0
Plane 6:  z=0, c=2, t=1    Plane 12: z=1, c=2, t=1

Implementation Modules

• iw_bioio.py: BioIO-based implementation for local files
• iw_ngff.py: OME-NGFF/Zarr implementation using ome-zarr-py
• iw_blitz.py: OMERO Blitz implementation for remote servers

All implementations produce identical hashes for identical pixel data, conforming to the IMAGEWALK specification (IEP-0018).

Command-Line Interface

biocode - Generate Biocode (ISCC-SUM)

Generate biocode (ISCC-SUM) for bioimage scenes:

iscc-bio biocode INPUT [OPTIONS]

Options:
  -s, --source [auto|bioio|omero|zarr]  Data source type
  --simprints                           Generate per-plane simprints
  --host TEXT                           OMERO server hostname
  --iid INTEGER                         OMERO image ID
  --fid INTEGER                         OMERO fileset ID

imagecode - Generate Imagecode (Experimental)

Create comprehensive bioimage fingerprints with ISCC-SUM + ISCC-IMAGE + ISCC-MIXED codes:

iscc-bio imagecode INPUT [OPTIONS]

Options:
  -o, --output-dir PATH    Save extracted view PNGs
  -n, --max-views INTEGER  Maximum views per scene (default: 5)

views - Extract Representative Views

Extract intelligent 2D views for perceptual hashing:

iscc-bio views INPUT [OPTIONS]

Options:
  -s, --strategies TEXT    View strategies (mip, best_focus, representative, composite)
  -n, --max-views INTEGER  Maximum views to extract (default: 8)
  -o, --output-dir PATH    Directory to save thumbnails
  --host TEXT              OMERO server hostname
  --iid INTEGER            OMERO image ID

scenes - Extract Scene Thumbnails

Extract thumbnails from all scenes in a multi-scene file:

iscc-bio scenes INPUT

thumb - Extract Thumbnail

Extract a single representative thumbnail from a bioimage:

iscc-bio thumb INPUT

Python API

High-Level Entry Point

The biocode() function handles all supported sources (local files, OME-NGFF/Zarr, OMERO) and returns a single container dict: a top-level raw-byte ISCC-SUM over the original source with the per-scene IMAGEWALK content codes nested as parts:

from iscc_bio.api import biocode

result = biocode("image.ome.tiff")
print(result["iscc_code"])             # top-level raw-byte ISCC-SUM
print(result["parts"][0]["iscc_code"])  # per-scene IMAGEWALK content code

# OME-NGFF/Zarr store (top-level code is the tree SUM over the store)
result = biocode("dataset.zarr")

# OMERO server (top-level iscc_code is None; parts carry per-scene codes)
result = biocode(host="omero.server.com", username="user", password="pass", iid=123)

IMAGEWALK Plane Iteration

from iscc_bio.imagewalk import iter_planes_bioio, iter_planes_ngff
from iscc_bio.imagewalk import iter_planes_blitz_image

# Iterate over planes using BioIO
for plane in iter_planes_bioio("image.czi"):
    print(f"Scene {plane.scene_idx}, Z={plane.z_depth}, "
          f"C={plane.c_channel}, T={plane.t_time}")
    print(f"Shape: {plane.xy_array.shape}, dtype: {plane.xy_array.dtype}")

# Iterate over OME-Zarr planes
for plane in iter_planes_ngff("data.zarr"):
    # Process plane.xy_array (2D numpy array)
    pass

# Iterate over OMERO planes
from omero.gateway import BlitzGateway
conn = BlitzGateway("user", "pass", host="omero.server.com")
conn.connect()
image = conn.getObject("Image", 123)

for plane in iter_planes_blitz_image(conn, image):
    # Process plane.xy_array
    pass
conn.close()

Generate Biocode (ISCC-SUM)

from iscc_bio.biocode import generate_biocode
from iscc_bio.imagewalk import iter_planes_bioio

# Generate biocode for all scenes
planes = iter_planes_bioio("image.lif")
results = generate_biocode(planes)
print(results[0]["iscc_code"])  # ISCC-SUM for first scene

Generate Imagecode (Experimental)

from iscc_bio.imagecode import generate_imagecode, format_output

# Generate comprehensive fingerprints (ISCC-SUM + ISCC-IMAGE + ISCC-MIXED)
fingerprints = generate_imagecode("image.nd2", max_views=5)

# Format output
output = format_output(fingerprints, "image.nd2")
print(output)

Supported Formats

Via BioIO plugin ecosystem:

• OME-TIFF/TIFF: Multi-page TIFF with OME-XML metadata
• OME-Zarr/NGFF: Next-generation file format
• OMERO: Remote server access via Blitz gateway
• CZI: Carl Zeiss Image format
• ND2: Nikon NIS-Elements format
• LIF: Leica Image File format
• DV: DeltaVision format
• BioFormats: 150+ formats via Bio-Formats Java library

Development

Setup Development Environment

# Clone repository
git clone https://github.com/bio-codes/iscc-bio.git
cd iscc-bio

# Install with all dependencies
uv sync --extra all

# Run CLI during development
uv run iscc-bio --help

Development Commands

# Run tests
uv run pytest

This project uses poethepoet for task automation:

# Format markdown files
uv run poe format-md

# Format code files
uv run poe format-code

# Build documentation
uv run poe docs-build

# Run all formatting and docs
uv run poe all

Architecture

Core Modules

• iscc_bio.api: High-level Python API — biocode() entry point for all sources

• iscc_bio.imagewalk: IMAGEWALK plane traversal implementations

  • iw_bioio.py: BioIO implementation
  • iw_ngff.py: OME-Zarr/NGFF implementation
  • iw_blitz.py: OMERO Blitz implementation
  • common.py: Plane data model and canonical byte conversion

• iscc_bio.biocode: Biocode (ISCC-SUM) generation from IMAGEWALK plane iterators

• iscc_bio.imagecode: Imagecode generation (ISCC-SUM + ISCC-IMAGE + ISCC-MIXED)

• iscc_bio.views: Intelligent view extraction strategies

• iscc_bio.cli: Command-line interface (thin wrapper around the Python API)

Design Principles

1. Lazy Loading: Uses Dask arrays for memory-efficient processing of large images
2. Format Agnostic: Identical processing logic across all formats via IMAGEWALK
3. Deterministic: Reproducible hashes across platforms and formats
4. Modular: Clean separation between traversal, canonicalization, and hashing

Funding

This work was supported through the Open Science Clusters’ Action for Research and Society (OSCARS) European project under grant agreement Nº101129751.

See: BIO-CODES project (Enhancing AI-Readiness of Bioimaging Data with Content-Based Identifiers).

License

Apache License 2.0 - See LICENSE file for details.

Citation

If you use iscc-bio in your research, please cite:

@software{iscc_bio,
  title        = {bio-codes/iscc-bio: ISCC Processing for Bioimage Data},
  author       = {Pan, Titusz},
  year         = 2025,
  url          = {https://github.com/bio-codes/iscc-bio},
  note         = {Supported by OSCARS (Open Science Clusters' Action for Research and Society) under European Commission grant agreement Nº101129751},
  version      = {0.1.0}
}

Related Projects

• iscc-sum - Fast ISCC Data-Code and Instance-Code hashing
• iscc-core - ISCC Core Algorithms - Reference implementation
• iscc-lib - ISCC Core Algorithms - Polyglot Rust implementation
• BioIO - Bioimage reading library
• OME-Zarr - Next-generation file format implementation